evidence that carotid artery intima-media thickness (IMT) is also associated with increased occurrence of CHD. Apolipoprotein E (ApoE) has been widely studied in regard to its role in lipid transport and metabolism, but the role that variation in the APOE gene plays in relation to carotid artery IMT and risk of incident CHD remains a subject of debate. The authors examined the effect of each APOE allele (e2, s3, s4) on LDL-C and carotid IMT,as well as the association with CHD risk, in 12,491 participants of the ARIC study. APOE s2, s3 and s4 relative allele frequencies were determined, respectively in European-Americans (0.08, 0.77,0.15)and African-Americans (0.11, 0.67,0.22). Using Cox regression, it was shown that these alleles did not predict incident CHD in either racial group. The APOE s2 allele was associated with lower LDL-C and the s4 allele with higher LDL-C in both European-Americans and African-Americans. The APOE s2 and s4 alleles were associated with carotid IMT measures in both racial groups, but after adjusting for lipid parameters, only the s4 allele was associated with carotid IMT measures in African-Americans. This study directly addresses a major question being addressed by Project 3: Will genetic variants that predict intermediate traits also predict endpoints? IV.C.4.2.2. Rare combinations of multiple common DMA sequence variations predict high risk subgroups. SNPs have been hypothesized to explain the genetic predisposition to CHD in the general population. Lack of evidence for a role of such variation is fostering pessimism about the utility of genetic information in the practice of medicine. In this study we determined the utility of exonic and 5'SNPs in APOE and lipoprotein lipase (LPL) when considered singly and in combination for predicting incidence of CHD in 8,456 individuals from the general population during 24 years of follow-up. In men,LPL D9N improved prediction of CHD (p=0.03) beyond smoking, diabetes and hypertension. The group of men heterozygous and homozygous for the rare D9N variant had a hazard ratio of 1.69 (95% confidence interval =1.10-2.58) relative to the most common genotype. Pairwise combinations of D9N with -219G>T in APOE and N291S and S447X in LPL significantly improved the prediction of CHD (p=0.05 in women, p=0.04 in men, p=0.03 in men, respectively) beyond smoking, diabetes and hypertension, and identified subgroups of individuals (n=6 to 94) with highly significant hazard ratios of 1.92to 4.35.These results were validated in a case-control study (n=8,806). This study establishes that combinations of SNPs in the APOE and LPL genes identify subgroups of individuals with substantially increased risk of CHDbeyond traditional risk factors. This result is consistent with the heterogeneity of the etiology of CHD and the hypothesis that different combinations of variations in susceptibility genes are predictors in different subsets of those with CHD. IV.C.4.2.3. Using PRIM to identify SNPs that reproducibly predict variation in disease risk beyond that afforded by established risk factors. All predictor variables are not expected to be present and relevant in all cases of disease. When population strata are not defined, PRIM107'109 is an alternative strategy to traditional logistic regression for identifying partitions of observations with increased prevalence that are also predictive of the risk for unstudied members of the population of inference. Multiple mutually exclusive partitions are produced, each grouping together observations with equivalent values of significant predictor variables. The predictor variables will vary from partition to partition. Details of PRIM and its application to AIMS 3.1-3.3 are presented in Section IV.C.3.4.2.4 page 174. The results of our experience in the application of this analytical strategy to data collected in a large population-based study of CHD events in Copenhagen, Denmark110 are summarized in Figure IV.C.4.2.3.1, page 179.The prevalence of disease in the sample of 5,455 individuals studied was 0.095. We identified six partitions using traditional risk factors (age,gender, smoking, diabetes and hypertension) with prevalence estimates ranging from 0.034 to 0.195. Five of these partitions were further PHS 398/2590 (Rev.09/04, Reissued 4/2006) Page 178 Continuation Format Page